C h a p t e r 3 4

Group Studies in Experimental Neuropsychology

Lesley K. Fellows

Why Neuropsychology? individual differences in skull shape (explicitly thought to be a proxy for underlying brain structure) A fundamental assumption of neuropsychology, and in relation to individual differences in behavior. of cognitive neuroscience more generally, is that Complex traits like benevolence and wit were thus behavior has a biological basis—that it results from related to particular parts of the brain. Although the processes that are executed in the nervous system. methods are clearly flawed to the eye of the modern Following from this assumption, emotions, reader, the underlying concept of localization, that thoughts, percepts, and actions can be understood brain structure and function are related, had a major in neurobiological terms. This premise was impact on the development of clinical neurology and advanced by the philosophers of ancient Greece, of experimental neuropsychology. supported, in part, by observations of patients with The work of Broca, Wernicke, and other 19th- brain injury (Gross, 1995). The fact that damage to and early 20th-century neurologists illustrated how the brain could lead to paralysis, disorders of sensa- observation in clinical populations can (a) provide tion, or even disruptions of consciousness suggested insights into how a complex behavior (like lan- that this organ was the seat of such abilities, guage) can be segmented into simpler components although the broad claim that brain function under- (e.g., production and comprehension) and (b) how lies behavior was not without controversy over the such components can be related to specific regions centuries that followed (Crivellato & Ribatti, 2007). of the brain. Both defining the components of The 19th century saw, on the one hand, major behavior and relating these components to the developments in understanding the anatomy and brain can be done on the basis of a single, carefully physiology of the brain and, on the other, more sys- studied case (see Chapter 33 of this volume). How- tematic descriptions of behavioral changes resulting ever, the limitations of clinical observations in from neurological diseases. These advances laid the humans were also apparent in these early days. Cli- groundwork for current thinking about the brain. nicians were (and are) acutely aware of wide vari- Here again, clinical observations provided an impor- ability in the clinical presentation of a particular tant impetus, as did analyses of individual differ- pathological condition, determined both by differ- ences in normal behavior: Neurologists such as Paul ences in premorbid individual characteristics Broca and Carl Wernicke reported that focal brain (e.g., age, education, or health status) and differ- injury to specific areas within the left hemisphere ences in the specific details of the pathological pro- disrupted particular aspects of language (Feinberg & cess. Case series and group studies provide an Farah, 2006). Their thinking was influenced, in part, important means of determining the generalizability by Franz-Joseph Gall and others who developed the of inferences that can be drawn from individual concept of phrenology in about the same period. observations. Phrenology was based on observations of specific

DOI: 10.1037/13620-034 APA Handbook of Research Methods in Psychology: Vol. 2. Research Designs, H. Cooper (Editor-in-Chief) 647 Copyright © 2012 by the American Psychological Association. All rights reserved. Lesley K. Fellows

Gordon Holmes, a British neurologist whose cognitive process (Fellows et al., 2005; Rorden & work on the effects of penetrating brain injury in Karnath, 2004). In principle, this is a powerful form World War I soldiers helped to establish the retino- of evidence because it addresses causality. Although topic organization of primary visual cortex, poeti- cognitive neuroscience now has many other meth- cally captured the limitations of clinical observation ods available to investigate brain–behavior relations, in a lecture delivered in 1944: most provide correlational data. Standard functional neuroimaging methods, for example, reveal brain My own work on the visual cortex has regions in which blood-oxygen-level dependent been limited to observation in man. . . . (commonly referred to as BOLD) signal (itself a cor- This has required the collection of a relate of neural activity) is correlated with a behav- large number of observations, for while ioral process of interest. These findings can be the physiologist can rely on experiments informative, but alone they are insufficient to estab- when he can select and control, . . . the lish that the brain regions so identified are in fact clinician must depend on the analysis of necessary for the behavior in question (Fellows et observations which are rarely so simple al., 2005; Rorden & Karnath, 2004). or clear cut.. . . The physiologist may be These inferential considerations are particularly compared with the builder in . . . hewn relevant in the study of complex behaviors, and in stones which can easily be fitted together, new areas of enquiry. Consider risky decision mak- the physician resembles the mason who ing as an example. Imagine yourself at the blackjack has to use irregular rubble and there- table, deciding how much to stake on the next card. fore requires more time and labour to Several correlated processes are likely under way in attain his end. But in some branches of your brain. You may be calculating the odds of win- neurology, the “rubble” collected and ning, integrating your recent history of wins and put together by the clinician is essential. losses, and weighing these factors to reach a deci- (Holmes, 1944/1979, pp. 440–441) sion. You may be imagining how you would spend Holmes underlined two key points: (a) that the your winnings, or how you would explain a loss to limitations inherent to studying the effects of brain your spouse. It is likely that you are experiencing injury in humans can be minimized by gathering substantial changes in arousal and autonomic tone: data from many subjects and by interpreting these A pounding heart and sweaty palms often accom- data in the context of converging evidence from pany a risky choice. Whether all of these putative other methods and (b) that the limitations are offset processes are distinct, important to the decision, or by the fact that these observations provide crucial simply correlated epiphenomena are empirical ques- insights that may not be acquired in any other way. tions. Interpreting functional magnetic resonance As this chapter will describe, there have been many imaging (fMRI) activations in this situation is not logistical, technical, and analytic advances in human easy—for example, is a given area more active lesion studies over the past century. However, because it is critically involved in risky decision ­Holmes’s comments on the core advantages and making or is it important in central autonomic con- ­limitations remain as pertinent as ever. trol, mediating the changes in sympathetic nervous system outflow that result in the pounding heart? Although careful design can help to minimize these Inferential Strengths of Lesion uncertainties of interpretation, the nature of correla- Studies tional evidence means that they can never be elimi- Research on effects of brain injury on behavior nated entirely. Converging evidence from addresses two main issues: First, it can establish that loss-of-function methods, such as lesion studies, can a particular region of the brain is necessary for the help test necessity claims. If we take a hypothetical expression of a particular behavior, in turn, support- “risky decision” brain area as an example, a study of ing the inference that it is critical for a particular patients with damage to that area could directly test

648 Group Studies in Experimental Neuropsychology whether it was critical for the decision, for the auto- exclude potential lesion-related confounds that nomic changes that accompany that decision, or for might explain the observations in a single case; for both. Such an experiment would shed light both on example, they can establish that it is the site of dam- the critical components of decision making (do age, rather than its etiology, that underlies the autonomic changes influence choice?) and on the behavior change. brain substrates of the critical processes (e.g., Another common and related problem that can Critchley et al., 2003). be addressed by group studies is that lesions are More profoundly, the study of patients can pro- often more extensive, or less precisely located, vide biological constraints to psychological theory. than is ideal for testing a given structure–function The usual form this has taken is that of dissociation hypothesis. If the function is disrupted but the of cognitive processes. Two putative psychological lesion is large, the conclusions cannot be specific. If constructs may be considered distinct if brain injury a group of patients with lesions varying in extent disrupts one and not the other—establishing what is but overlapping in some smaller area are found to termed a function dissociation. As will be described, have a common impairment in function, one can experiments of this kind have been influential, but infer that the function likely relies on the region of how these are best designed and interpreted is not overlap that is common across patients. Recent without controversy. methods have built on this logic to allow statistical tests of structure–function relations at the voxel level and will be discussed in more detail later in What Do Group Studies Add to this chapter. Thus, lesion extent can limit structure– the Analysis of Single Cases? function mapping in single case studies but can at Group studies address two potential problems of least be addressed, and maybe turned to advantage, interpretation that plague single cases: One is that in group studies. observed deficits in a single patient may be due to Group studies in neuropsychology are strictly premorbid differences in function—normal individ- observational rather than experimental. Like case- ual differences may thus be misattributed to the control studies in epidemiology, they are vulnerable lesion. This may be implausible for some deficits: to confounds and biases, but they can nevertheless Common sense dictates that major hemiparesis or offer important insights. These biases are predictable visual field defects are outside the range of normal and generally can be avoided with careful design or variation and can generally be safely linked to the addressed with appropriate analyses. The observa- brain injury. But other aspects of behavior, such as tional nature of this approach to lesion–function executive functions, emotional, or social processes, mapping means that there is no imposed direction- may differ substantially across healthy individuals, ality: Studies may begin from either lesion or func- making it more likely that such a difference will be tion. These two perspectives are discussed in turn. found by chance in a brain-injured patient. Idiosyn- crasies in brain organization, in structure–function Designs Driven by Behavior mapping, or in recovery from brain injury can also contribute to exceptional performance in a A major challenge in both psychology and cognitive single case. neuroscience is to define the architecture of behav- Even if we can safely assume that the patient’s ior. One way or another, the complexity of behavior brain, function, and brain–function relations were needs to be parsed into analyzable constituents, representative before the injury, a second source of whether these are conceptualized as modules, pro- variability would make group studies important. For cesses, or interacting networks (Dunn & Kirsner, obvious reasons, brain lesions in human subjects are 2003). The challenge is to identify the appropriate not under experimental control. As a result, there is constituent parts and then to understand how they substantial inherent variability in the extent and interact from both a psychological and a neural causes of brain injury. Group studies help to point of view.

649 Lesley K. Fellows

Arguably, this enterprise has been most success- How Do You Know a Dissociation ful when it has been closely linked to neurobiology. When You See One? For example, we now have a detailed understanding The logic of dissociation is clear in principle but can of visual processing that begins from response prop- be challenging to operationalize in practice. (See erties of single neurons in the retina, moves to how Dunn & Kirsner, 2003, for a more detailed analysis these are combined in the initial stages of cortical of these challenges.) How intact must a group be in visual processing, and continues from there to the Task A? How impaired in Task B? What is the likeli- computations that support object or face recognition hood of such dissociations occurring by chance, in a (Farah, 2004; Van Essen, Felleman, DeYoe, Olavar- given population, and for any given pair of tasks? ria, & Knierim, 1990). This enterprise obviously One common approach is to test for a crossover requires data gathered with a variety of methods. interaction in the performance of two tasks, across Studies in patients with brain injury can provide two groups, but other patterns may be as or more important insights into the biologically relevant important, depending on the relations between the lines of cleavage for a given (complex) behavior, by tasks and on the relations between a given cognitive helping to identify associations and dissociations process and performance on the task that is meant between putative component processes. to measure it (Bates, Appelbaum, Salcedo, Saygin, & When behavior is treated as the independent Pizzamiglio, 2003; Dunn & Kirsner, 2003; Shallice, variable, patients are selected on the basis of the 1988). presence of some behavioral manifestation—either a clinical syndrome or performance on a particular Conceptual Precision task. Additional behavioral measures aiming to iso- An important first step in any experiment of this late putative component processes are then adminis- sort is to start from a position of conceptual clarity: tered to determine whether these processes are, in A model of the component processes of interest that fact, distinct (i.e., dissociable). A single dissociation is well-justified will dictate the appropriate analyses. refers to a situation in which subjects are impaired A priori hypotheses might come from existing on a task that presumably assesses a particular abil- experimental work in humans using lesion or other ity but are unimpaired on another task that assesses methods, from animal studies, from computational a separate ability. Single dissociations are evidence models, or from combinations of these sources. in favor of a hypothesis that the tasks measure dis- tinct component processes (Damasio & Damasio, Measurement Reliability 1989; Shallice, 1988). However, there are practical Once processes of interest are identified, tasks are issues that make alternative explanations for such needed to measure the relevant behavior as specifi- patterns quite likely: As one example, dissociations cally as possible. Ideally, such measures will have assume that the tasks being used are approximately good psychometric properties: no ceiling or floor equally difficult. An easy task and a hard task tap- performance, good test–retest reliability, and perfor- ping the same component process would show mance that is minimally influenced by demographic apparent dissociation because at least some patients or education factors (Laws, 2005). Brain-injured would fail the hard task but pass the easy task patients are typically older and less educated, on (Shallice, 1988). average, than the convenience samples of healthy This potential explanation is less likely if a dou- undergraduates often used in the development of ble dissociation can be demonstrated: Here, one set new measures. Because the time and energy of these of patients fails Task A but does well on Task B, patients are limited, it is wise to pilot new tasks in whereas another set shows the opposite pattern. healthy subjects who are otherwise demographically The explanatory power and experimental elegance similar to the target patient population. That said, of double dissociation has been a touchstone since the appropriate reference population for the actual the early days of experimental neuropsychology experiment may not be healthy subjects. Depending (Teuber, 1955). on the hypothesis, patients with brain injury may

650 Group Studies in Experimental Neuropsychology provide more relevant comparison data, and such healthy populations. People differ in their cognitive comparisons may be less affected by the ceiling capacities. This may be particularly true for certain effects that can be a problem in healthy reference cognitive capacities. This variation is generally only groups. a nuisance in lesion studies, increasing the variance Measurement variation, that is, the extent to across both experimental and reference groups. It which task performance will vary if the same sub- can become a confound, however, if such individual ject is tested repeatedly, is a source of noise that in differences are not randomly distributed across principle is under the experimenter’s control. It groups. This can be due to sampling error, system- may have important influences on the analysis and atic sampling bias, or nonindependence of lesion- should be minimized to the extent possible (Bates, related variables. The simplest form of sampling Appelbaum, et al., 2003). In addition to piloting in error is that, by chance, more subjects from one end demographically relevant healthy populations, of the normal range are present in one group than attention needs to be given to particular challenges another. This risk is minimized by increasing the that may arise in patient populations. Depending sample size (indeed, avoiding this risk is an impor- on the patient population of interest, relevant tant motivation of group versus single case studies), issues might include (a) difficulty understanding but patient studies have practical limits on sample instructions, (b) difficulty with motor or percep- size that make this a challenge. tual aspects of the tasks that are related to the Sampling bias may occur for other reasons. For lesion but not of interest (e.g., because of weakness example, some normal individual difference may interfering with responding, or disruption of pri- also make it more likely that an individual would mary sensory processing), and (c) nonspecific suffer a particular neurological injury. This problem changes in arousal or attention related to the injury is illustrated, for example, in studying the links or to psychoactive medications (e.g., anticonvul- between impulsivity and the frontal lobes: If patients sants) that may be more commonly taken by those who have suffered frontal lobe damage from trau- in the target group than in the reference group. matic brain injury are found to be more impulsive, Some of these issues can be addressed in the exper- does this establish that the frontal lobes are impor- iment. For example, patients may need simplified tant in impulse control? Or do impulsive people get instructions, additional practice, or modifications into situations in which they suffer such injuries of how stimuli are presented or responses col- more often than the less impulsive, so that this nor- lected. These problems also apply to single case mal individual difference ends up overrepresented studies, but their solutions may be different in in the patient group? group studies. In single cases, there may be more A second source of individual variability relates flexibility in optimizing the details of the task to to inevitable variation in the nature and extent of accommodate patient-specific factors. In groups, brain injury within a group. Furthermore, variation there is a trade-off between using a consistent mea- in lesion location is not the only lesion-related sure across all participants (and so allowing the determinant of this kind of variability: Factors such results to be easily pooled) and adapting the task to as comorbidity and medication use may differ sys- individual restrictions. tematically with lesion etiology or location and so be another source of bias. Interindividual Variability A second source of variability relates not to the mea- Do Lesion Data Matter in Behavior-Driven surement tools but rather to the individuals being Designs? measured. Individual differences in group lesion One can test the hypothesis that two processes are studies can be conceptualized as arising from three functionally dissociable in a patient population potential sources, and these differences can be of no without ever considering the details of their lesions. interest or of major interest. The first source is indi- In principle, dissociation, particularly double disso- vidual differences of the same sort that one finds in ciation, addresses the issue. In practice, however,

651 Lesley K. Fellows there are many nuances in determining what the imaging. This is less resource intensive, minimizes thresholds might be for establishing dissociations, patient inconvenience, and often provides lesion data including the need to consider departures from cor- that are of more than adequate resolution for testing relations across tasks as well as (or instead of) abso- a given hypothesis. Regardless of approach, the qual- lute performance in each of two tasks. Even if the ity (and so anatomical precision) of the lesion char- experimental goal is purely to understand the archi- acterization needs to be considered in the analysis. tecture of cognitive processes, rather than their rela- The simplest way of presenting lesion data is to tion to the brain, lesion analysis can provide reproduce the imaging as-is for each patient. This external validation of claims of dissociation. Consis- works well for single cases but becomes awkward in tent lesion location–function mapping bolsters the group studies. Indeed, it is only appropriate to argument that what impaired (or unimpaired) reproduce individual scans if the behavioral data are patients have in common is disruption of a specific also presented for each individual, that is, in case system, rather than some demographic or task- series format. If behavioral data are presented as related confound (Robertson, Knight, Rafal, & group means, imaging data also need to be pre- Shimamura, 1993). sented in a form that allows insights into what is common in the group. This can be achieved simply—for example, by tabulating the number of Lesion-Driven Designs subjects with damage to particular regions. How- It is equally possible to study structure–function ever, modern imaging data are acquired in digital relations in the human brain with the brain injury form, permitting group lesion data to be presented treated as the independent variable. Rather than as brain images that are much more accessible, and aiming to discern how behavior can be dissected, also are more easily related to the wider literature, the starting point is to determine the cognitive pro- particularly fMRI studies. cesses for which a given brain region is necessary. Individual lesions first need to be represented in Of course, these two aims converge on the same a common space. This can be achieved in two main central questions. ways: either by manually tracing the lesion onto some common template (Damasio & Damasio, Characterizing Lesions 1989; Kimberg, Coslett, & Schwartz, 2007) or by In the early days of neuropsychology, lesion charac- manually or automatically defining the lesion on the terization was based on neurosurgical sketches, plain individual patient’s anatomical scan and then warp- X-rays of the skull, or the results of autopsies. Com- ing the brain (and the lesion) onto a standard tem- puterized tomography (CT) and, more recently, plate. The first method is labor intensive and structural magnetic resonance imaging (MRI), have requires substantial expertise. The second method dramatically improved the quality of anatomical data. relies on the same algorithms used to warp individ- The first step in characterizing lesions is thus to ual scans into common space for fMRI analysis in acquire either MRI or CT images of each patient’s healthy subjects and can be more automatized. brain. MRI is preferred because it offers better resolu- However, the anatomical distortions caused by the tion, and in many cases better sensitivity, than CT, presence of a lesion lead to particular technical and it avoids exposing the participant to ionizing issues that need to be addressed thoughtfully if this radiation. However, MRI may be contraindicated in approach is taken (Nachev, Coulthard, Jager, Ken- patients with pacemakers or surgical clips, for exam- nard, & Husain, 2008; Rorden & Brett, 2000). ple, or not tolerated because of claustrophobia. Ide- Regardless of approach, defining the boundaries of ally, high-resolution imaging should be acquired in lesions always involves some judgment and so is a the whole patient sample using standard parameters potential source of error. and equipment, as close to the time of behavioral Registering individual lesions to a common tem- testing as possible. That said, it may be much more plate allows these data to be shown in aggregate— practical to use the most recently available clinical most commonly as overlap images (generated by

652 Group Studies in Experimental Neuropsychology representing the arithmetic sum of damage in each conclude that the effects are due to damage in a par- voxel, across the group), which show the degree ticular region. They may relate to some effect of to which damage affects common brain structures brain damage more generally, including effects of for a given group of patients (Frank, Damasio, & confounding factors that may be more common in Grabowski, 1997; Makale et al., 2002; Rorden & those with brain injury than in those without. Fur- Brett, 2000). Such images can also demonstrate the thermore, it may be difficult to avoid ceiling effects absence of common damage in two groups that are in a healthy control group. To address these prob- meant to be anatomically distinct. Digitized lesion lems, many studies include a comparison group data that are represented in a common space can be with brain injury that spares the region of interest. If used in more complex computations, including sta- the aim is to exclude generic effects of brain damage tistical tests of structure–function relations (see the (or confounds more likely to be present in ill than in section Finer Grained Lesion–Symptom Mapping), healthy participants) in the interpretation of the and are more readily linked to other sources of data findings, then any site of damage that spares the that are also expressed in common brain coordi- region in question is fine. This is something of a nates, notably fMRI studies. missed opportunity, however. If the lesioned com- parison group is selected so that the lesions affect a second, specific brain region, then that group can Region of Interest Designs serve double duty: both controlling for nonspecific When there is an a priori hypothesis about the func- effects of injury, and establishing that the other tional role of a particular brain region, region of region is not necessary for the process in question. interest (ROI) designs are appropriate. Here, partici- Such a targeted approach also assesses the possibil- pants are identified on the basis of the presence of ity that the lesioned reference is impaired on some damage affecting (or sometimes restricted to) some other task, providing insurance against the claim specified region of the brain. Behaviors of interest that the reference group is somehow less impaired are measured with one or several tasks, and perfor- for whatever reason. In the end, one is left with a mance is compared with appropriate reference focused double dissociation. groups. If impairment is identified, it is evidence This elegant design is not easy to achieve and, that the brain region plays a necessary role in task when achieved, is still potentially susceptible to the performance and, by extension, in the cognitive problems described for functional dissociations. The process of interest. The major advantage of this main practical difficulty is in recruiting an appropri- approach is its hypothesis-driven design and the ate lesioned comparison group, whether it involves statistical power that accompanies such designs. patients with anatomically common or disparate This power means that relatively small sample sizes lesions, matched to the experimental group on both may be adequate, particularly because effect sizes in clinical and demographic variables. Systematic lesion studies are often quite large. Such designs recruitment methods, such as patient registries, can may have directional hypotheses, making one-tailed make this more feasible (Fellows, Stark, Berg, & statistical tests appropriate. Chatterjee, 2008). There are several important design issues to con- If it is impossible to recruit a lesioned compari- sider in these focused studies. The first is the appro- son group, then the next best approach is to thor- priate reference group. One common approach is to oughly characterize the relation between compare participants with damage to a particular demographic variables and task performance in a region to a healthy group, matched on demographic healthy reference group, in which adequate sample characteristics. This provides some control over sizes are much more feasible, and then use that potential demographic confounds (although perhaps information to inform analyses of the patient data. not as much as one might think, depending on the For example, demographic characteristics that differ sample size and the variance in these demographic between patient groups can be shown not to sub- characteristics). However, one cannot unequivocally stantially influence task performance in a large

653 Lesley K. Fellows healthy reference group, or the influence can be Pitfalls in ROI Designs characterized sufficiently to allow these contribu- Recruitment Bias tions to be covaried out in the primary analysis. A The observational nature of human lesion studies, in common approach along these lines is to express general, requires particular care to minimize poten- performance of each lesioned participant as a per- tial bias. When testing a structure–function hypoth- centile or z score on the basis of performance in a esis with an anatomical ROI design, efforts should larger healthy reference sample (e.g., Gläscher et al., be made to include all subjects with damage to the 2009; Tsuchida & Fellows, 2009). Although desir- region in question. It is common to undertake what able, this may not always be feasible, depending on might be called a hybrid study—for example, includ- the reference data available for a given task. ing patients with both left hemisphere damage and It is important to consider what an ROI design aphasia and then asking a more specific question does not do: It does not necessarily impose a true ana- about language processes. This runs the risk of dis- tomic boundary. Also, it is obvious that nothing will torting the results. At the least, it may magnify be learned about the potential contributions of brain apparent structure–function relations, by picking regions outside that boundary. Perhaps less obvi- desired patients with both lesion and dysfunction. It ously, there can be a risk of not detecting effects that may give spurious findings as well because subjects are, in fact, related to damage within the boundary. without impairment provide important constraints This can happen if the region is much larger than the in lesion–function mapping (Rorden, Fridriksson, & actually critical brain area; effects caused by damage Karnath, 2009). At the other end of the spectrum, in the smaller area are diluted by normal performance patients may be excluded because they are too in those with damage affecting the larger, but not crit- impaired to perform the experimental tasks. This is ical, area. The group as a whole will have variable per- unavoidable but important to keep in mind. For formance, and the statistical analyses may fail to example, it might be difficult to study the neural detect effects. Finally, effects that are detected with a processes related to the control of behavior. If dam- given ROI may nevertheless have been better captured age to some key structure resulted in severe agita- by a different anatomical boundary. tion, such patients are unlikely to be approached (and if approached, to agree) to participate in cogni- Statistics for ROI Designs tive neuroscience research. Similarly, severe aphasia often precludes informed consent, so such patients When the data are considered as group means, the may be systematically excluded. same statistical approaches used for comparing groups in any study are appropriate. ROI designs commonly are limited to small samples and may Control Groups involve skewed behavioral data (either because of As in epidemiologic case-control studies, the refer- ceiling effects in the control group, or substantial ence group is important in lesion studies. In princi- variability in the patient group, or both). These ple, those in the control group should differ from issues obviously need to be taken into account when the patient group only in that they have not suffered planning the analysis, if they cannot be avoided in a brain injury. Subjects drawn from the same popu- the design. Sometimes group studies are better ana- lation are optimal. Practically, this is often challeng- lyzed as a series of single cases. This approach may ing to achieve, but it can be accomplished in several be suitable when the group of patients varies widely ways. If healthy subjects are needed, then individu- on relevant demographic or other variables, or als with similar demographic profiles should be indeed, in task performance. Sometimes this recruited, perhaps even friends or family members approach is taken post hoc, in which case, the of the patients. Patients with damage to other brain results should be considered with particular caution, regions (but due to the same causes as damage in given the ease with which confounds other than the group of interest) are often better choices lesion location may explain observed effects. because they are more likely to be matched on

654 Group Studies in Experimental Neuropsychology potentially confounding variables, such as medica- effects of stroke evolve over time (Fruhmann Berger, tion usage, or nonspecific psychological effects of Johannsen, & Karnath, 2008). The effects of brain serious illness. However, depending on the size of damage can be studied at any time point, but this is the groups and the anatomical specificity of the a highly relevant variable and must be considered in hypotheses, it can be harder to match such subjects both study design and interpretation. The develop- on demographic variables. ment of MRI sequences that can delineate ischemic brain tissue very shortly after the onset of acute All Lesions Are Not Created Equal stroke has provided the opportunity to do hyper- Lesions do not occur at random. There are system- acute lesion–function mapping (Marsh & Hillis, atic biases in who suffers a brain injury, in the 2008; Newhart, Ken, Kleinman, Heidler-Gary, & extent to which an injury that affects one part of the Hillis, 2007). Such work can identify regions that brain will be accompanied by damage to other parts are normally necessary for a given function. In con- of the brain, in the destructiveness of a given injury, trast, studies examining chronic impairments after and in the time course and mechanisms of recovery brain damage are perhaps better thought of as iden- from that injury. For example, ischemic stroke dam- tifying regions that are necessary for the recovery of ages parts of the brain that are supplied by particular a given function: Deficits still present months or blood vessels. These vascular territories mean that years after an injury are, by definition, resistant to damage to one area, for example, inferior frontal compensatory mechanisms. lobe, will be more commonly associated with dam- age to another area in the same territory (e.g., Finer Grained Lesion–Symptom Mapping insula, inferior parietal lobe). Conversely, such Regions of interest can, in principle, be any size. In damage will almost never be associated with damage practice, there is a lower limit of resolution imposed to the areas that are supplied by other blood vessels, by the volume of brain tissue that is injured in indi- such as the other hemisphere, the frontal pole, or vidual subjects, the extent to which those volumes the occipital lobe. Furthermore, some vascular terri- overlap in a given sample, or the resolution of the tories are more commonly affected than others. imaging methods that are used to characterize the Injury to the areas supplied by the middle cerebral injury. Lesion volume, rather than imaging resolu- artery, for example, will be overrepresented in a tion, is typically the limiting factor in the MRI era. given series of unselected stroke patients. These reg- The upper limit of resolution is determined by con- ularities have implications for interpreting the ceptual issues; determining that some function is results of lesion studies (Rorden & Karnath, 2004) related to the integrity of the whole brain, for exam- and constrain the brain regions that can be readily ple, is likely to be of limited interest. That said, studied by lesion methods. many core concepts in neuropsychology began with Lesion etiology can also affect the accuracy of regions of interest encompassing entire cerebral the lesion mapping, and the observable structure– hemispheres, and defining such broad structure– function relations: Slow-growing tumors push nor- function relations may still be important as cogni- mal brain tissue aside without necessarily disrupting tive neuroscience tackles new areas of study, such as function, which can lead to lesions that appear quite in social or affective domains. large but have much milder functional effects. Corti- Converging evidence argues that structure– cal resections in epilepsy have precise margins and function relations are considerably more discrete spare the white matter—two advantages—but the than is captured by examining hemispheric, or even cortex that is resected is often not normal and may lobar, effects. There are practical limits to the have been abnormal for a long time. regional specificity that can be attained in group Relatedly, the degree to which compensation can studies with ROI designs. If the study is restricted to occur depends on the time course over which the patients with damage to some specific and small brain disorder develops, its extent, and the time brain area, an adequate sample is unlikely to be since injury. There is no doubt that the functional recruited in a reasonable time. An alternative is to

655 Lesley K. Fellows enroll patients with variable damage to a relatively limitations of sample size, is in determining how to broad region—even one hemisphere or the whole appropriately correct for multiple comparisons. brain—and then undertake analyses to establish Recently, statistical methods that were developed which subregion contributes to the observed deficits for fMRI have been adapted for examining structure– in function. dysfunction relations at a voxel-by-voxel level. This There are three main approaches to analyzing is a natural extension of multi-ROI designs, with the data from patients who have variable damage in a (potential) advantage of principled control of multi- large brain region. The one with the longest history ple comparisons. Once lesions volumes are regis- involves a secondary analysis in a standard ROI tered to a common template, univariate statistics study. Having established that some anatomically can be applied to test whether the performance of defined group is impaired and observing the usual patients with damage to a given voxel differs from variability in that impairment, one may ask whether performance of patients with damage that spares the there is an anatomical basis to that variability. That voxel. This results in a statistical map showing the is, whether damage to a specific subregion is a main strength of association between damage and dys- determinant of task performance. This can be function in anatomical space. addressed qualitatively by examining the pattern of This approach, commonly referred to as voxel- lesions in the impaired and unimpaired subgroups, based lesion–symptom mapping (VLSM), does not in essence, carrying out a behavior-driven analysis require imposing potentially arbitrary (or somehow nested in the original ROI study. Lesion overlap “wrong”) ROI boundaries and allows task perfor- methods are often used to this end: The lesion over- mance to be considered either as a dichotomous lap image for impaired and unimpaired subgroups (intact–impaired) or continuous variable. The latter can be examined visually, or lesion extent can be avoids having to impose a second, potentially arbi- subtracted across these groups to identify the poten- trary, boundary on the data. VLSM also has the tially critical subregion (e.g., Milne & Grafman, potential to map networks, that is, to identify several 2001). Alternative but analogous methods include regions that may contribute to task performance tabulating the presence or absence of injury to Brod- within a single experiment. Several variations of this mann areas and comparing the outcome in patients method, using different statistical approaches, have with and without behavioral impairment. been developed (see Bates, Wilson, et al., 2003; Chen, There are drawbacks to this approach. First, it is Hillis, Pawlak, & Herskovits, 2008; Kinkingnéhun et important to realize that it is usually undertaken al., 2007; Rorden et al., 2009; Rorden & Karnath, post hoc. Any finding needs confirmation in a new 2004; Rorden, Karnath, & Bonilha, 2007; Solomon, experiment that is designed to test the specific ROI a Raymont, Braun, Butman, & Grafman, 2007). priori. Selection bias and confounding factors can These advantages come with trade-offs. As with easily influence the results. Such subregion analyses fMRI analysis, this massively univariate approach usually involve very small sample sizes, and it can requires conservative correction for multiple com- be impossible to properly account for other (e.g., parisons, which in turn demands a substantial sam- demographic) contributors to observed effects. This ple size. The number of subjects is not the only approach is also prone to problems because of the consideration; lesion overlap and distribution are nonindependence of damage (see the section All also important determinants of a study’s power. Lesions Are Not Created Equal). Results from such Methods exist to estimate the anatomical extent of analyses should be treated with particular caution adequate power in a given sample, and this is an when the a priori, ROI-based analysis does not important adjunct in interpreting VLSM analyses establish significant differences between groups. A (Kimberg et al., 2007; Rudrauf et al., 2008). System- multiple ROI approach can be applied a priori. Sev- atic approaches to patient recruitment are also criti- eral studies have taken this approach (e.g., Picton cal in acquiring a suitable sample size, and in et al., 2007; Stuss, Murphy, Binns, & Alexander, ensuring that the sample has been appropriately 2003). The main difficulty, beyond the perennial characterized (Fellows et al., 2008).

656 Group Studies in Experimental Neuropsychology

White Matter Damage and Disconnection cognitive functions that may be necessary for a Effects given task are also degraded in more or less corre- With the exception of certain neurosurgical resec- lated ways. tions, lesions are rarely confined to a single struc- ture and often disrupt the white matter leading into Conclusion or away from a given gray matter region or fibers of passage (i.e., adjacent tracts that may have nothing Studies of disrupted function can provide important to do with the damaged gray matter beyond physical insights into the architecture of cognitive processes proximity). This can pose challenges in interpreting and can identify the brain substrates critical for lesion studies. Observed behavioral effects might be these processes. Lesion studies in humans have par- due to the white matter damage, which would be ticular inferential strengths, explaining their long particularly misleading if it involves fibers of pas- and fruitful history in neuroscience and psychology, sage. Modern neuroimaging can assess the extent of and recent advances in anatomical imaging and sta- white matter injury, either in standard structural tistical analysis contribute to the continued rele- scans, or by using tract-specific imaging such as dif- vance of such work (Catani & ffytche, 2010; fusion tensor imaging. Furthermore, white matter Chatterjee, 2005). The ability to learn about brain atlases are becoming increasingly sophisticated. function from the experience of people with brain Thus, methods exist to address possible white mat- injury has intrinsic worth beyond its inferential ter contributions and are beginning to be applied to logic: Patients and families can provide rich descrip- structure–function mapping (Catani, Jones, & tions of how brain damage has affected their lives, ffytche, 2005; Karnath, Rorden, & Ticini, 2009; which can lead to unexpected insights beyond the Philippi, Mehta, Grabowski, Adolphs, & Rudrauf, laboratory context. Such anecdotal evidence can 2009; Rudrauf, Mehta, & Grabowski, 2008; Thie- provide interesting starting points for hypothesis- baut de Schotten et al., 2008; Urbanski et al., 2008). driven experiments, the results of which may in turn Developments in image analysis to study net- be directly relevant to patient care. The observa- work properties of the brain, whether captured by tional nature of these studies requires thoughtful structural or functional measures, may prove useful experimental design, and this chapter has aimed at as adjuncts to the lesion approaches discussed so far providing an overview of the main factors to be con- (Dosenbach, Fair, Cohen, Schlaggar, & Petersen, sidered in such designs. 2008; He, Dagher, et al., 2009; He, Wang, et al., 2009). At the least, these techniques draw attention References to network-oriented conceptual frameworks. Bates, E., Appelbaum, M., Salcedo, J., Saygin, A. P., & Pizzamiglio, L. (2003). Quantifying dissociations Clinical Conditions With Diffuse Damage in neuropsychological research. Journal of Clinical and Experimental Neuropsychology, 25, 1128–1153. Brain–behavior relations can also be studied in clini- doi:10.1076/jcen.25.8.1128.16724 cal conditions that have multifocal or diffuse dam- Bates, E., Wilson, S. M., Saygin, A. P., Dick, F., Sereno, age. Traumatic brain injury, multiple sclerosis, and M. I., Knight, R. T., & Dronkers, N. F. (2003). degenerative dementias are examples. Imaging Voxel-based lesion-symptom mapping. Nature methods can quantify regional cortical and white Neuroscience, 6, 448–450. matter changes, even when these are subtle or dif- Catani, M., & ffytche, D. H. (2010). On “the study of the fuse, and such changes can be correlated with nervous system and behaviour.” Cortex, 46, 106–109. doi:10.1016/j.cortex.2009.03.012 behavior. Most of the pitfalls that have been dis- cussed also apply to such studies. There are addi- Catani, M., Jones, D. K., & ffytche, D. H. (2005). Perisylvian language networks of the human tional challenges in interpreting anatomical data brain. Annals of Neurology, 57, 8–16. doi:10.1002/ when multiple areas are dysfunctional in a more or ana.20319 less correlated (and more or less detectable) way, Chatterjee, A. (2005). A madness to the methods and in interpreting behavioral data when multiple in cognitive neuroscience? Journal of Cognitive

657 Lesley K. Fellows

Neuroscience, 17, 847–849. doi:10.1162/089892 Neuron, 61, 681–691. doi:10.1016/j.neuron.2009. 9054021085 01.026 Chen, R., Hillis, A. E., Pawlak, M., & Herskovits, E. H. Gross, C. G. (1995). Aristotle on the brain. Neuroscientist, (2008). Voxelwise Bayesian lesion-deficit analy- 1, 245–250. doi:10.1177/107385849500100408 sis. NeuroImage, 40, 1633–1642. doi:10.1016/ He, Y., Dagher, A., Chen, Z., Charil, A., Zijdenbos, A., j.neuroimage.2008.01.014 Worsley, K., & Evans, A. (2009). Impaired small- Critchley, H. D., Mathias, C. J., Josephs, O., O’Doherty, world efficiency in structural cortical networks in J., Zanini, S., Dewar, B. K., . . . Dolan, R. J. (2003). multiple sclerosis associated with white matter lesion Human cingulate cortex and autonomic control: load. Brain: A Journal of Neurology, 132, 3366–3379. Converging neuroimaging and clinical evidence. doi:10.1093/brain/awp089 Brain: A Journal of Neurology, 126, 2139–2152. He, Y., Wang, J., Wang, L., Chen, Z. J., Yan, C., Yang, H., doi:10.1093/brain/awg216 . . . Evans, A. (2009). Uncovering intrinsic modu- Crivellato, E., & Ribatti, D. (2007). Soul, mind, brain: lar organization of spontaneous brain activity in Greek philosophy and the birth of neuroscience. humans. PLoS ONE, 4(4), e5226. doi:10.1371/ Brain Research Bulletin, 71, 327–336. doi:10.1016/ journal.pone.0005226 j.brainresbull.2006.09.020 Holmes, G. (1979). The organization of the visual cortex in Damasio, H., & Damasio, A. R. (1989). Lesion analysis in man. In C. G. Phillips (Ed.), Selected papers of Gordon neuropsychology. New York, NY: Oxford University Holmes (pp. 438–451). Oxford, England: Oxford Press. University Press. (Original work published 1944) Dosenbach, N. U., Fair, D. A., Cohen, A. L., Schlaggar, Karnath, H. O., Rorden, C., & Ticini, L. F. (2009). B. L., & Petersen, S. E. (2008). A dual-networks Damage to white matter fiber tracts in acute spa- architecture of top-down control. Trends in Cognitive tial neglect. Cerebral Cortex (New York, NY), 19, Sciences, 12, 99–105. doi:10.1016/j.tics.2008.01.001 2331–2337. doi:10.1093/cercor/bhn250 Dunn, J. C., & Kirsner, K. (2003). What can we Kimberg, D. Y., Coslett, H. B., & Schwartz, M. F. (2007). infer from double dissociations? Cortex, 39, 1–7. Power in voxel-based lesion-symptom mapping. doi:10.1016/S0010-9452(08)70070-4 Journal of Cognitive Neuroscience, 19, 1067–1080. Farah, M. J. (2004). Visual agnosia (2nd ed.). Cambridge, doi:10.1162/jocn.2007.19.7.1067 MA: MIT Press. Kinkingnéhun, S., Volle, E., Pelegrini-Issac, M., Golmard, Feinberg, T. E., & Farah, M. (2006). A historical perspec- J. L., Lehericy, S., du Boisgueheneuc, F., . . . Dubois, tive on cognitive neuroscience. In T. E. Feinberg & B. (2007). A novel approach to clinical-radiological M. Farah (Eds.), Patient-based approaches to cognitive correlations: Anatomo-Clinical Overlapping Maps neuroscience (pp. 3–20). Cambridge, MA: MIT Press. (AnaCOM): Method and validation. NeuroImage, 37, 1237–1249. doi:10.1016/j.neuroimage.2007.06.027 Fellows, L. K., Heberlein, A. S., Morales, D. A., Shivde, G., Waller, S., & Wu, D. H. (2005). Method matters: Laws, K. R. (2005). “Illusions of normality”: A method- An empirical study of impact in cognitive neurosci- ological critique of category-specific naming.Cortex, ence. Journal of Cognitive Neuroscience, 17, 850–858. 41, 842–851. doi:10.1162/0898929054021139 Makale, M., Solomon, J., Patronas, N. J., Danek, A., Fellows, L. K., Stark, M., Berg, A., & Chatterjee, A. Butman, J. A., & Grafman, J. (2002). Quantification (2008). Establishing patient registries for cognitive of brain lesions using interactive automated soft- neuroscience research: Advantages, challenges, and ware. Behavior Research Methods, Instruments, and practical advice based on the experience at two cen- Computers, 34, 6–18. doi:10.3758/BF03195419 ters. Journal of Cognitive Neuroscience, 20, 1107–1113. Marsh, E. B., & Hillis, A. E. (2008). Dissociation between doi:10.1162/jocn.2008.20065 egocentric and allocentric visuospatial and tactile Frank, R. J., Damasio, H., & Grabowski, T. J. (1997). neglect in acute stroke. Cortex, 44, 1215–1220. Brainvox: An interactive, multimodal visualization doi:10.1016/j.cortex.2006.02.002 and analysis system for neuroanatomical imaging. Milne, E., & Grafman, J. (2001). Ventromedial prefrontal NeuroImage, 5, 13–30. doi:10.1006/nimg.1996.0250 cortex lesions in humans eliminate implicit gender Fruhmann Berger, M., Johannsen, L., & Karnath, H. stereotyping. Journal of Neuroscience, 21(12), RC150. O. (2008). Time course of eye and head deviation Nachev, P., Coulthard, E., Jager, H. R., Kennard, C., & in spatial neglect. Neuropsychology, 22, 697–702. Husain, M. (2008). Enantiomorphic normalization of doi:10.1037/a0013351 focally lesioned brains. NeuroImage, 39, 1215–1226. Gläscher, J., Tranel, D., Paul, L. K., Rudrauf, D., Rorden, doi:10.1016/j.neuroimage.2007.10.002 C., Hornaday, A., . . . Adolphs, R. (2009). Lesion Newhart, M., Ken, L., Kleinman, J. T., Heidler-Gary, J., mapping of cognitive abilities linked to intelligence. & Hillis, A. E. (2007). Neural networks essential

658 Group Studies in Experimental Neuropsychology

for naming and word comprehension. Cognitive incorporation in group-level lesion studies. Cortex, and Behavioral Neurology, 20, 25–30. doi:10.1097/ 44, 1084–1096. doi:10.1016/j.cortex.2008.05.005 WNN.0b013e31802dc4a7 Shallice, T. (1988). From neuropsychology to mental struc- Philippi, C. L., Mehta, S., Grabowski, T., Adolphs, R., ture. New York, NY: Cambridge University Press. & Rudrauf, D. (2009). Damage to association fiber doi:10.1017/CBO9780511526817 tracts impairs recognition of the facial expression of emotion. Journal of Neuroscience, 29, 15089–15099. Solomon, J., Raymont, V., Braun, A., Butman, J. A., doi:10.1523/JNEUROSCI.0796-09.2009 & Grafman, J. (2007). User-friendly software for the analysis of brain lesions (ABLe). Computer Picton, T. W., Stuss, D. T., Alexander, M. P., Shallice, Methods and Programs in Biomedicine, 86, 245–254. T., Binns, M. A., & Gillingham, S. (2007). Effects of doi:10.1016/j.cmpb.2007.02.006 focal frontal lesions on response inhibition. Cerebral Cortex (New York, NY), 17, 826–838. doi:10.1093/ Stuss, D. T., Murphy, K. J., Binns, M. A., & Alexander, cercor/bhk031 M. P. (2003). Staying on the job: The frontal lobes control individual performance variability. Brain: A Robertson, L. C., Knight, R. T., Rafal, R., & Shimamura, Journal of Neurology, 126, 2363–2380. doi:10.1093/ A. P. (1993). Cognitive neuropsychology is more brain/awg237 than single-case studies. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, Teuber, H. L. (1955). Physiological psychology. Annual 710–717. doi:10.1037/0278-7393.19.3.710 Review of Psychology, 6, 267–296. doi:10.1146/ annurev.ps.06.020155.001411 Rorden, C., & Brett, M. (2000). Stereotaxic display of brain lesions. Behavioural Neurology, 12, 191–200. Thiebaut de Schotten, M., Kinkingnéhun, S., Delmaire, C., Lehericy, S., Duffau, H., Thivard, L., . . . Rorden, C., Fridriksson, J., & Karnath, H. O. (2009). An Bartolomeo, P. (2008). Visualization of disconnec- evaluation of traditional and novel tools for lesion tion syndromes in humans. Cortex, 44, 1097–1103. behavior mapping. NeuroImage, 44, 1355–1362. doi:10.1016/j.cortex.2008.02.003 doi:10.1016/j.neuroimage.2008.09.031 Tsuchida, A., & Fellows, L. K. (2009). Lesion evidence Rorden, C., & Karnath, H. O. (2004). Using human brain that two distinct regions within prefrontal cortex are lesions to infer function: A relic from a past era in the critical for n-back performance in humans. Journal of fMRI age? Nature Reviews Neuroscience, 5, 812–819. Cognitive Neuroscience, 21, 2263–2275. doi:10.1162/ doi:10.1038/nrn1521 jocn.2008.21172 Rorden, C., Karnath, H. O., & Bonilha, L. (2007). Urbanski, M., Thiebaut de Schotten, M., Rodrigo, S., Improving lesion-symptom mapping. Journal of Catani, M., Oppenheim, C., Touze, E., . . . Bartolomeo, Cognitive Neuroscience, 19, 1081–1088. doi:10.1162/ P. (2008). Brain networks of spatial awareness: jocn.2007.19.7.1081 Evidence from diffusion tensor imaging tractography. Rudrauf, D., Mehta, S., Bruss, J., Tranel, D., Damasio, H., Journal of Neurology, Neurosurgery, and Psychiatry, 79, & Grabowski, T. J. (2008). Thresholding lesion over- 598–601. doi:10.1136/jnnp.2007.126276 lap difference maps: Application to category-related Van Essen, D. C., Felleman, D. J., DeYoe, E. A., Olavarria, naming and recognition deficits.NeuroImage , 41, J., & Knierim, J. (1990). Modular and hierarchi- 970–984. doi:10.1016/j.neuroimage.2007.12.033 cal organization of extrastriate visual cortex in the Rudrauf, D., Mehta, S., & Grabowski, T. J. (2008). macaque monkey. Cold Spring Harbor Symposia on Disconnection’s renaissance takes shape: Formal Quantitative Biology, 55, 679–696.

659